Is Attention Interpretable?. Serrano, S. & Smith, N. A. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 2931–2951, Stroudsburg, PA, USA, 2019. Association for Computational Linguistics. ![Is Attention Interpretable? [link]](https://bibbase.org/img/filetypes/link.svg "link")
Paper doi abstract bibtex 1 download Attention mechanisms have recently boosted performance on a range of NLP tasks. Because attention layers explicitly weight input components' representations, it is also often assumed that attention can be used to identify information that models found important (e.g., specific contextualized word tokens). We test whether that assumption holds by manipulating attention weights in already-trained text classification models and analyzing the resulting differences in their predictions. While we observe some ways in which higher attention weights correlate with greater impact on model predictions, we also find many ways in which this does not hold, i.e., where gradient-based rankings of attention weights better predict their effects than their magnitudes. We conclude that while attention noisily predicts input components' overall importance to a model, it is by no means a fail-safe indicator.
@inproceedings{Serrano2019,
abstract = {Attention mechanisms have recently boosted performance on a range of NLP tasks. Because attention layers explicitly weight input components' representations, it is also often assumed that attention can be used to identify information that models found important (e.g., specific contextualized word tokens). We test whether that assumption holds by manipulating attention weights in already-trained text classification models and analyzing the resulting differences in their predictions. While we observe some ways in which higher attention weights correlate with greater impact on model predictions, we also find many ways in which this does not hold, i.e., where gradient-based rankings of attention weights better predict their effects than their magnitudes. We conclude that while attention noisily predicts input components' overall importance to a model, it is by no means a fail-safe indicator.},
address = {Stroudsburg, PA, USA},
archivePrefix = {arXiv},
arxivId = {1906.03731},
author = {Serrano, Sofia and Smith, Noah A.},
booktitle = {Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics},
doi = {10.18653/v1/P19-1282},
eprint = {1906.03731},
file = {:Users/shanest/Documents/Library/Serrano, Smith/Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics/Serrano, Smith - 2019 - Is Attention Interpretable.pdf:pdf},
keywords = {method: attention},
pages = {2931--2951},
publisher = {Association for Computational Linguistics},
title = {{Is Attention Interpretable?}},
url = {https://www.aclweb.org/anthology/P19-1282},
year = {2019}
}
Downloads: 1
{"_id":"urcxNWR79jaMCBWHD","bibbaseid":"serrano-smith-isattentioninterpretable-2019","authorIDs":[],"author_short":["Serrano, S.","Smith, N. A."],"bibdata":{"bibtype":"inproceedings","type":"inproceedings","abstract":"Attention mechanisms have recently boosted performance on a range of NLP tasks. Because attention layers explicitly weight input components' representations, it is also often assumed that attention can be used to identify information that models found important (e.g., specific contextualized word tokens). We test whether that assumption holds by manipulating attention weights in already-trained text classification models and analyzing the resulting differences in their predictions. While we observe some ways in which higher attention weights correlate with greater impact on model predictions, we also find many ways in which this does not hold, i.e., where gradient-based rankings of attention weights better predict their effects than their magnitudes. We conclude that while attention noisily predicts input components' overall importance to a model, it is by no means a fail-safe indicator.","address":"Stroudsburg, PA, USA","archiveprefix":"arXiv","arxivid":"1906.03731","author":[{"propositions":[],"lastnames":["Serrano"],"firstnames":["Sofia"],"suffixes":[]},{"propositions":[],"lastnames":["Smith"],"firstnames":["Noah","A."],"suffixes":[]}],"booktitle":"Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics","doi":"10.18653/v1/P19-1282","eprint":"1906.03731","file":":Users/shanest/Documents/Library/Serrano, Smith/Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics/Serrano, Smith - 2019 - Is Attention Interpretable.pdf:pdf","keywords":"method: attention","pages":"2931–2951","publisher":"Association for Computational Linguistics","title":"Is Attention Interpretable?","url":"https://www.aclweb.org/anthology/P19-1282","year":"2019","bibtex":"@inproceedings{Serrano2019,\nabstract = {Attention mechanisms have recently boosted performance on a range of NLP tasks. Because attention layers explicitly weight input components' representations, it is also often assumed that attention can be used to identify information that models found important (e.g., specific contextualized word tokens). We test whether that assumption holds by manipulating attention weights in already-trained text classification models and analyzing the resulting differences in their predictions. While we observe some ways in which higher attention weights correlate with greater impact on model predictions, we also find many ways in which this does not hold, i.e., where gradient-based rankings of attention weights better predict their effects than their magnitudes. We conclude that while attention noisily predicts input components' overall importance to a model, it is by no means a fail-safe indicator.},\naddress = {Stroudsburg, PA, USA},\narchivePrefix = {arXiv},\narxivId = {1906.03731},\nauthor = {Serrano, Sofia and Smith, Noah A.},\nbooktitle = {Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics},\ndoi = {10.18653/v1/P19-1282},\neprint = {1906.03731},\nfile = {:Users/shanest/Documents/Library/Serrano, Smith/Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics/Serrano, Smith - 2019 - Is Attention Interpretable.pdf:pdf},\nkeywords = {method: attention},\npages = {2931--2951},\npublisher = {Association for Computational Linguistics},\ntitle = {{Is Attention Interpretable?}},\nurl = {https://www.aclweb.org/anthology/P19-1282},\nyear = {2019}\n}\n","author_short":["Serrano, S.","Smith, N. A."],"key":"Serrano2019","id":"Serrano2019","bibbaseid":"serrano-smith-isattentioninterpretable-2019","role":"author","urls":{"Paper":"https://www.aclweb.org/anthology/P19-1282"},"keyword":["method: attention"],"metadata":{"authorlinks":{}},"downloads":1},"bibtype":"inproceedings","biburl":"https://www.shane.st/teaching/575/win20/MachineLearning-interpretability.bib","creationDate":"2020-01-09T19:34:05.106Z","downloads":1,"keywords":["method: attention"],"search_terms":["attention","interpretable","serrano","smith"],"title":"Is Attention Interpretable?","year":2019,"dataSources":["okYcdTpf4JJ2zkj7A"]}